I’m sorry, Dave, I’m afraid I can’t do that

I’ve had a quick ‘play’ with Artificial Intelligence previously but I decided to take the plunge and get myself the relevant accounts set up so I could use some of the better AI image creation tools and see what all the fuss is about. Is AI something to be worried about, a useful tool or just utter crap?

The main contenders at the minute appear to be DALL-E, from Open AI, the creators of Chat GPT that you must all of heard of by now, Midjourney and Leonardo AI. They all do a similar thing – enter a description of what you want on their web page, and out pops a few images for you to choose, refine, etc. They do this with varying degree’s of success/failure.

I discounted DALL-E as it is the least capable of the bunch right now, and decided to concentrate on Midjourney and Leonardo. Both require you to first create an account in Discord and then from their website link back to your Discord account. From Discord you have a link to each of there servers and the interface for each is essentially a text prompt – you just describe the image you want. This is not quite correct, while Leonardo doesn’t insist on a Discord account it kind of does, and only gains you access to the prompt on the Leonardo site once you have one, but once you have access it is through the Leonardo site, but is very similar. Midjourney is all done through Discord. Please bear in mind unless you subscribe to one of the premium packages for these services the images you create cannot be used commercially and EVERYONE can see both the prompt you used to create the image and the output, which raises a lot of questions about intellectual property rights, etc. On the top tier service from Midjourney you can go into stealth mode and nobody can see what you are doing, but there are still some legal issues to be worked out on how you can use these images as AI is so new.  Watch this space with regards to image rights, usage, etc develops.

Anyway I digress. I wanted to check both platforms with a simple prompt: –

carrot on a white plate on a red and white checked table cloth

Here’s the results from Leonardo AI

There is a carrot on a plate, but it’s all a bit weird and there are extra things in the images that just don’t look right.

Here’s the results from Midjourney with the exact same prompt

Much better results, but you can tell it has learned from other images that food has cutlery also on the plate and put it there without asking. A better description may resolve this and a description of the lighting may also help.

Assuming the software has learned what certain objects and products look like I decided to create an image from a commercial product. An Audi TT.

an audi TT RS in orange with black wheels driving fast down a mountain road. moody atmospheric lighting 4k

Here’s the images from Leonardo AI

These could not be used, they are far too rough and kind of what I have seen from DALL-E. Although I have seen much better images from Leonardo when it is used for gaming style content as that is its target audience. The results I saw from Midjourney were very different however

I have to ask the question is it blending stock images with images it creates here to get the result? I decided to blend real-world with fantasy to see what it came up with. I only did this on Midjourney I was almost out of free credits on Leonardo and wanted to save them for now.

This is the prompt I used next

imagine a new car that is a cross between an audi tt rs and the batmobile. Painted electric blue, driving through a neon city scape with bright lights in the rain at night. 4k

It is starting to get a bit more interesting and you could see how you could play with this idea and refine it, but I still think it is using stock images for a lot of what it ‘knows’.

To check the randomness and make sure it wasn’t relying solely on stock images I asked one of my friends (yes I do have some) to come up with the next prompt. Here’s what he came up with.

a luner lander picking its nose sitting by pile of green cats

Completely random, and here’s the results

It understands lunar lander and a green cat, but not anything about picking ones nose. To prove a point, here’s an image of ‘boy picking nose’

It doesn’t really get that at all. If I try ‘boy with finger in nose’ this is a bit more successful, but ultimately unsuccessful.

There’s also a lot of similarities between the image of the ‘boy’, as if there is a limited set of material it has learned from.  Somethings clearly aren’t there yet, but what about something more descriptive that we see every day. Dinner.

dinner table with a plate of chicken madras and pilau rice served on a white plate with a gold ornate pattern. There is more pilau rice in a white and gold bowl and more of the curry in another bowl. On another plate is a steaming garlic naan bread. The table is rustic wood and there is a large glass of cold beer. The walls of the room are decorated in indian colours and the lighting is warm, and comfortable but not too bright. The cutlery is ornate and gold.

Here’s the images from Midjourney

It missed the beer for a start, there’s no gold on the plates and there’s no gold cutlery. Let’s see what Leonardo can do.

Better and worse. It got the beer, but the rest of it is a bit off and there are some weird artefacts in the images. There is clearly a reliance on stock images here.

Let’s try a simple one – Elvis riding a Horse.

If you squint your eyes a bit that could be Elvis riding a horse. Let’s try a more challenging one.

Elvis riding a horse through vegas on a night, neon lights, cyberpunk style

Here’s Midjourney

And here’s Leonardo

It’s easy to see this is just a tool for now as neither produces a set of images that look like Elvis.  It is a very powerful tool if used correctly or incorrectly for that matter – I’ve just seen someone use this as a prompt

8k realistic photo, natural lighting, imploded submarine next to recked titanic ship submerged underwater, –v 5.1 –ar 16:9

If we take something more familiar and something we as buzzers might do ourselves we can see how useful this could potentially be

bluebells in a wooded glade appearing through a shaft of light

Here’s Leonardo’s version

And finally Midjourney

Both look quite good, but on some of the Midjourney images I noticed something strange – look at the bottom right. It’s obviously ripping the images off from someone as it learns. Ooopsie.

Both have their uses, and how good they are depends on you really. How well can you describe what you want. Leonardo, I believe, allows you to upload images to form the basis of your renders but I haven’t tried this yet. Personally I think AI is just another tool for now, although the rate of development could see this change in the very near future.  I set Midjourney a task to create a line drawing of one of our dogs so Charlotte could colour it in.

colouring pave for kids, cute pugalier puppy wagging tail in a park, cartoon style, thick lines, low detail, no shading.

And here’s a picture of Noah, just in case you’ve not seen a Pugalier before. It’s not a million miles off.

Now this could quite easily be imported into Illustrator and be adapted as the basis for an illustrated book.  It gives a good starting point and only took a few seconds to do.  This is the real power of image AI for now – it’s an ideas and prototyping tool.

I would recommend everyone should have a play with Midjourney and Leonardo AI, or even DALL E just to see what they can do for you.  Also play with ChatGPT, it’s a very good tool to create ideas, etc.  I used it for keyword creation when I couldn’t think of any keywords.  Now there’s a thought, ask ChatGPT to create the prompts for Midjourney.  What images would AI create using AI?  Let’s test that.

I asked Chat GPT to creat an image of Elvis riding a horse it has to be photorealistic, he’s what it came out with

Create a photorealistic image of Elvis Presley confidently riding a majestic horse through a picturesque landscape. Capture the essence of Elvis’s charisma and style, showcasing him in a timeless moment of his journey. Pay attention to intricate details, such as his iconic hairstyle, distinctive facial features, and signature attire, while ensuring that the horse is equally captivating, radiating strength and elegance. Transport viewers to a bygone era, where Elvis and his equine companion embark on a remarkable midjourney together. Let your artistic prowess shine as you bring this extraordinary vision to life.

And here’s the results in Midjourney

It produced better images than I did, but you can still see Midjourney is nicking images from others.  Shit is going to fly over this at some point.

AI is here to stay and will become an increasing part of our lives as it is build into more and more products we use, such as the Adobe Suite, Bing or Microsoft Office.    Give it a go.

Published in Member Blogs
  1. Shaun, that’s really interesting. I hadn’t given any thought to how they source the images, but yes it’s obvious when you think about it. if you have Adobe Creative Cloud you can download the Photoshop Beta which has an AI generator built-in and you can then use it to modify your own photos using layers. That could be quite nifty (plus you’re not tied to credits).

    • I can see me using AI to automate tasks to make my workflow quicker. I don’t think I will be using it to create the final image as such. If you look at the way IT is going now, we have something called ‘hardware defined software’, which is where we create the hardware first and then the software to fit it later. For many years we have had a model where we have been software led and then improved the hardware so it can run. the software. The opposite is now true and this is the model Apple have adopted with M1 and companies like Nvidia are now adopting – the rest of the industry is playing catch up. When we take this hardware defined software approach we have lots of clever things programmers can use to access the hardware (video encoders, machine learning, etc) but without have to re-invent the wheel every time they do so. The result is AI built into Photoshop, etc to do some very clever things and do them very quickly too. Take this idea and put it on steroids and this is what Nvidia are doing in the data centres the are running the likes of ChatGPT. There has been this gradual shift away from Intel, and now we can all see why. A tad geeky, but this is why we are seeing all this AI stuff right now – fundamental changes in computer hardware.

  2. Interesting – I tried Bing AI on one of your text strings and got this…….

    • Cool. I knew Bing had ChatGPT built into it, but didn’t know how far it had gotten. That’s really useful to know

      • It produced some pretty good images using your text strings

        • Ask bing ‘show me a prompt to create an image of an xxxx using Bing as the image creator’ or something like that. Basically ask the AI what it would do to create the image. Like putting google into google. It produces better results that we would on our own. At the minute we aren’t very good at prompting AI, we need AI to prompt AI

Leave a Reply